Forecasting the Economic Effects of AI
AI-generated research notes, not an Unjournal evaluation. No person has checked these analyses or their numbers. The two outputs retain their own dates, methods and limitations.
Karger, Kuusela and colleagues elicit economic forecasts from economists, AI-industry and policy professionals, highly accurate forecasters and the public under specified AI-capability scenarios. This workspace brings together our earlier AI-assisted assessment and a later, blind run of Joe Francis’s isitcredible review suite.
What holds up, and what needs checking?
The clearest contribution is a structured description of the recruited respondents’ beliefs. Those forecasts have not been validated against future economic outcomes. Both analyses flag how broad capability scenarios, respondent selection and aggregation affect what readers can infer.
The later isitcredible run supports substantial disagreement conditional on the scenarios, while narrowing the interpretation that this identifies economic mechanisms as the primary source of ordinary disagreement. It also distinguishes fewer labor-force participants from AI-caused job losses. These are conclusions of an AI review, not findings checked by a person.
Check the detailed review
Joe Francis’s suite, run with Codex on October 29–30: claim-by-claim review, challenge, audit and digest. Browse 15 issues by type or severity, or search the evidence.
Read the earlier working assessment
Dated 2026-10-04; its substantive assessment date remains 2026-10-04. It covers recruitment, scenario framing, aggregation, downstream use and questions for the authors. It also contains provisional AI ratings.
The report is retained with its original content and dates. Integration into this workspace is not a new evidence check.
How the two analyses relate
The later isitcredible run was completed blind to the earlier working assessment. This website links the outputs after the run; it does not turn agreement between AI analyses into independent empirical validation.
Use the earlier report for broader methodological and prioritization context. Use the later digest for its audited issue ledger and narrowly scoped claim conclusions. Where wording differs, inspect the source and scope rather than treating this overview as a resolution.
What an independent evaluation could add
Check the substantive interpretations against the paper; examine recruitment and nonresponse, weighting, coherence interventions, distribution fitting and the disagreement decomposition. Native survey re-estimation still requires respondent-level data and analysis materials that this run did not obtain.
Paper prioritization and discussion · Paper PDF · Authors’ forecast summaries